Haswell graphics expectations: A quantum leap or just another small step forward?
As the Ivy Bridge processors from Intel comes out four months from now, its graphics, with reportedly double the Sandy Bridge GPU performance, would have an impact on entry level discrete GPUs. Will Haswell continue with the same major performance jump, or just some moderate improvements?
By now, we know that Ivy Bridge graphics, integrated in the CPU, will be quite a refresh compared to the current Sandy Bridge capabilities. The performance, DX11 compatibility, and high resolution multi-display capability should satisfy most mainstream users' needs, somewhat reducing the entry-level discrete GPU market at that point.
A year later, in early 2013, the next generation Haswell platform should see the light of the day, again first in the mainstream PC and mobile arena. Haswell will again include grapghics enhancements, but how far are those likely to go? Here we share some insights on the graphics side of Intel's next major X86 PC microarchitecture revision.
The 'Generation 7.5' graphics core in Haswell is, as the naming convention hints, a 'point upgrade' of the seventh generation Ivy Bridge graphics engine. Up to 20 execution units, each 128 bits wide, are to be supported, and these are supposedly enhanced with a new GPU compute architecture. The other major additions include a new internal GPU cache, hierarchical Z, as well as improved encode & decode pipelines for more hardware formats.
A very interesting productivity enhancement is the processor's ability to simultaneously manage external display connection both by its own integrated graphics and the external PCIe GPU over up to two displays, avoiding the need for external switching hardware.
How about the performance? It seems that the initial performance expectations aren't that high – compared to each matching Ivy Bridge graphics level, we're talking of maybe 20% average performance improvement. Two possible explanations come to mind, one being that Intel team is happy enough with the Ivy Bridge graphics speed-up, and focuses instead on features and power saving in Haswell (remember the power-focus on Haswell from their IDF keynote?), while the other one is that the actual speed-up is to be way higher, especially in the high-graphics GT3 version, and that future driver improvements should help as well.
We'll look more into the possible additional features and capabilities expected from Haswell generation graphics over time, however we also have to keep in mind that graphics is one capability where AMD has very strong credentials, including integrating it on the CPU. If, say, Llano graphics is roughly 2x that of Sandy Bridge (i.e. similar performance as Ivy Bridge), yet Haswell is another 20% or 70%, depending on the version, that would roughly match the performance of AMD 'Trinity' graphics, which is, however, due much earlier – at about the same time as the Ivy Bridge itself. On the other hand, AMD is unlikely to have a Fusion integration with a Bulldozer follow-on on the high end desktop yet by early 2013, so the only competition that would beat Haswell at the time in the GPU department is another mainstream follow-on to 'Trinity', maybe in 20 nm GloFo process?